DRV8353R: DRV fault at increased VM

We use the IPT020N10N3ATMA1 MOSFET

Dear Matt,

I have lowered the IDRIVE to 150mA src & 200mA snk and the issue persists.

It seems that playing around with the external RCD-clamp values does improve matters. Do you have a suggestion or best practice regarding clamps used with the DRV8353?

Hi Alexander,

We normally use RC snubbers without the diode (see this FAQ LINK). Snubber values should be tuned to the system. Is the behavior the same without the diodes?

Thanks,

Matt

We use a very thick copper foil on our PCB and a big drain cooling area. It's very hard to get the diodes off.

May I ask why you recommend not using them?

Hi Alexander,

It would temporarily hold the output low when trying to switch the phase from low to high. I'm not sure that is the root cause but it is not something we normally implement.

Can you provide an oscilioscope plot of a correct waveform and incorrect waveform? Put one channel on nFAULT and trigger on the falling edge. I'd like to see:

• These to determine if there is a particular channel dependency:
  • nFAULT, SHA, SHB, SHC
  • nFAULT, GHA, GHB, GHC
  • nFAULT, GLA, GLB, GLC
• If there is a channel dependancy, look at just the channel where the failure is observed
  • nFAULT, GHx, SHx, and GLx.

Thanks,

Matt

Hi Matt, I haven't done this yet but I have captured the waveforms (DS, GS, GGND of the hi-side and lo-side FETs) of 3 differently configured PCBs. Please consider:

PCB1 (1000mA src/2000mA snk, untuned snubber):

We can drive the motor reliably at a lowered Vsupp (30-32V). Alternatively if we start the motor at a lowered VSupp and increase this to 48-56V no issues occur, although sometimes the FAULT-condition still triggers. The waveforms were captured using this method.
If we start the motor at 48-55V the FAULT-condition triggers instantly. 

Hi-side MOSFET VDS:

Hi-side MOSFET VGS:

Hi-side MOSFET VG - GND:

Lo-side MOSFET VDS:

Lo-side MOSFET VGS:

Lo-side MOSFET VG - GND:

PCB2 (450mA src/900mA snk, untuned snubber):

We can drive the motor reliably at a lowered Vsupp (30-32V). Alternatively if we start the motor at a lowered VSupp and increase this to 48-56V no issues occur, although sometimes the FAULT-condition still triggers. The waveforms were captured using this method.
If we start the motor at 48-55V the FAULT-condition triggers instantly. 

Hi-side MOSFET VDS:

Hi-side MOSFET VGS:

Hi-side MOSFET VG - GND:

Lo-side MOSFET VDS:

Lo-side MOSFET VGS:

Lo-side MOSFET VG - GND:

PCB3 (450mA src/900mA snk, tuned snubber 23nF & 330R):

We can drive the motor reliably at the required Vsupp (48-55V) without triggering a FAUL. The waveforms were captured using this method.
Nonetheless the waveforms do not look that much different from the previous ones IMO.

Hi-side MOSFET VDS:

Hi-side MOSFET VGS:

Hi-side MOSFET VG - GND:

Lo-side MOSFET VDS:

Lo-side MOSFET VGS:

Lo-side MOSFET VG - GND:

Sorry in advance for the overwhelming number of pictures but perhaps you could make something out if this data.

Hi Alexander,

It's hard to see what is going on with only one trace at a time, can you take the captures with multiple traces and the fault pin on one screen so I can tell which shots were failing and which were not?

Thanks,

Matt

Your high-side MOSFET Vgs in PCB1 (1000mA src/2000mA snk, untuned snubber) & PCB2 (450mA src/900mA snk, untuned snubber) showcase quite a lot of ringing. I assume this is when the fault occurs?

Thanks,

Matt

Today I will measure this alongside the FAULT pin and tell you for sure.

Hi Alexander,

Sounds good. Please let me know what you measure!

Thanks,

Matt

Hi Matt!

We have captured some interesting waveforms which I think you will find interesting as well. Phases A & C look normal, but look what's going on when we take a look at B:

SHB (purple) & nFAULT (blue): 

GHB (purple) & nFAULT (blue):

GHB (green), SHB (purple) & nFAULT (blue):

It appears that after a certain number of PWM-pulses (this number seems random) the phase B GH takes almost 2us to charge. These plots are after cutting the GHB trace and replacing it with a wire. Before this experiment the waveform looked idetentical.

This capture shows the rising edge of a PWM period which did not trigger the nFAULT and where the rising edge is fast and clean. This is with the same PCB and same motor measuring the gate of the same B hi-side MOSFET.

Very interested in your take on this!

Alexander,

Matt and Michael are both out of office today, Let me ping our DRV8 person and they will respond early next week. Sorry for the delay.

Regards,

-Adam

Hi Adam, thank you very much! We're all very curious about the cause.

A funny observation: If we disconnect phase B from the PCB we still get the same error, even though the waveforms look clean enough. 

Hi Alexander,

Thank you for the additional information! Matt and Michael are out of the office today but will try to reply in the next couple of days. Thank you for your patience!

Regards,

Anthony 

Something thing else to consider, perhaps: an older version of the PCB works well despite featuring the exact same circuit with the MOSFET being the only difference. We use the FDBL86062-F085 instead of the IPT020N10N3ATMA1.

Hi Alexander,

Thanks for posting all of the waveform pictures. I apologize about the delay getting back to you. Several of the team members were out of the office.

Indeed, very interesting problem you're having.

To summarize:

- IDRIVE was lowered, however issue still persisted. 

- No issue driving the motor reliably at a lowered Vsupp (30-32V). If  motor is started at a lowered VSupp and increased to 48-56V no issues occur.

- Starting motor at 48-55V, the fault occurs instantly. 

- Phase A and C switch fine consistently and faults never occurs, even at 56V.

- The problem lies in phase B. GHB switches fine for some period of time, after that time a fault occurs. The fault occurs when the gate takes longer to slew, approximately 2us. I see that the fault occurs the moment GHB reaches its peak voltage. 

- On the 'good' waveform of GHB, we see the gate turn on <<1us.

The FETS that you're currently using are the FDBL86062-F085 correct?

If so, these FETS have a Qgd of 20nC. Try to aim for a risetime of 100ns-200ns. This would mean lowering your IDRIVE between 100mA src - 200mA src. 

Try lowering the IDRIVE one more time please. If this doesn't solve the issue, we can discuses further steps.

Thanks,

Michael

Dear Michael,

Thanks for getting back at me! Correct, except for the FET. We use the comparable IPT020N10N3ATMA1 MOSFET (QGD = 27nC). I have stated this earlier in one of my posts, you might have missed it.

We have tried all IDRIVE settings without any luck, tried playing around with the clamps, sometimes it seems to improve matters slightly, other times it doesnt.

The motor does kind of work at 48V at 1% PWM duty and spins veeeery slowly, but as soon as we switch to current control everything instantly stops. Also disconnecting the hi-side clamp and leaving the lo-side on does seem to help somewhat: in current control no faults are triggered at an asked current of 0A (which did not even work before) but as soon as that value increases - instant fault.

I don't know if this is allowed, but I would like to link my ee stack-exchange post since there is some additional info and captures there. Can I?

Hi Alexander,

Yes please link the ee stack thread, I will look through it.

Tomorrow I have a meeting with the team to review this and we will get back to you with our findings.

Thanks for your patience.

Regards,

Michael

Hi Alexander,

Looking at 6-th waveform from the top (Lo-side MOSFET VGS: for 1000/2000mA) we can see that it takes around 8us to fully discharge LS Mosfet gate charge. After around 4us from start of turning off LS Mosfet spike can be seen, caused probably by HS Mosfet turning on. That spike can cause turning LS mosfet on and crossconduction what may cause severe problems. Peak current gate drive time is 4us (plus 100ns of dead time) what would explain above mentioned timing.

Is there anything like resistors,ferrite beads between the following points of diagram?

GHX and GATE_HX,

SHX and PhaseX,

GLX and GATE LX.

Regards,

Grzegorz

Hello, Michael!

Thanks, and here you go!

Dear Grzegorz, 

No, the connections are uninterrupted.
But also, yes, this spike something we noticed as well when we started our captures.

Kr,

Alexander

Hi Alexander,

Thanks for your answer. Problems might be related to current value, if you have current probe, could you post waveform of output current together with nFault for a few fault cases. PCB layout may cause problems, driver IC may by faulty (it may be worth to replace it with a new one). I would also verify if VGLS and VCP voltages behave well when fault occurs.

Regards,

Grzegorz

Hi Grzegorz,

Sadly I do not have a current probe.

I doubt it's the IC, however, because we have this problem on multiple boards (we haven't tested all of them yet, though) It could be a faulty batch. 
I have not found the batch number on the IC (or where to look for it in the datasheet), perhaps we could test one from a different batch to eliminate this possibility? Do you know where to look for it or how to figure this out?

What should I look for on the VCP and VGLS? These do not fall below their UVP when the fault triggers. I do not have a capture of this, but I can make one if it helps you.

Kr, 

Alexander

Grzegorz,

If it's the layout, how would I go about verifying that it's causing issues? Because again, the captures look fine (by me, at least)...

Kr,

Alexander

Alexander,

If you experience problems on a few boards then faulty IC does'nt seem to be a probable cause. In case of IC batch I can not help you, I am just a guest here. Problems occur when much too long turn off/on Mosfet happens and I would just check if VCP and VGLS does not fall or rise by more than lets say 1V when fault happens.

PCB layout is the second most common problem (just after IDRIVE settings) in case of high current and high speed Mosfet drives. The problem lays in voltage spikes and ringing caused by high di/dt and parasitic inductances of pcb traces. The higher the current the higher di/dt, when you start motor at lower voltage motor current may be lower than when you start motor at higher voltage, this is why I suspect current value might be related to the problems. To verify that possibility I would accelerate motor at 50V slowly and see if problem still occurs.

In case of PCB layout you can read https://www.ti.com/lit/an/slva959a/slva959a.pdf  for a good start, share pcb layout for a review or sent it privately to TI engineer for a review.

Cheap probe like https://www.amazon.com/Hantek%C2%AE-Current-Clamp-kHz-20mA/dp/B06W2KFZLW could be a great help during motor drives design process.

Regards,

Grzegorz

Dear Grzegorz,

Regarding your idea: when in current control we have no issues on the lower voltages and cannot even start the motor at 48-55V. So starting it slowly would not be possible. I will share the layout of this board and the previous iteration here:

MOSFET drain-areas on the top-layer:

MOSFET drain and source areas and DRV cooiling areas on inner 1: 

MOSFET drain and source areas on the inner 2: 

This is the gnd-layer and my most controversial decision. I needed to increase the cooling areas underneath the FETS. As you can see in the captures, the signals and waveforms look ok despite this.

The layout of the half-bridges on the bottom layer: 

The layout of the DRV8353RH on the bottom layer: 

A 3D-view as reference: 

The half-bridge layout of a previous, working, HW version:

The DRV layout of a previous, working, HW version:

MOSFET drain and source areas and DRV cooiling areas on inner 1 of the working HW version: 

3D-view of the previous, working, HW-revision as reference:

Hi Alexander,

Thanks for all above pictures. First I will list all possible design flaws that I can find on first 6 pictures (not working board I assume).

1. PhaseX/SHX traces (Inner layer 1) go from LS Mosfet drains to right side of PCB and seem to be very long -> high parasitic inductance 

2. Current sense resistors (Bottom layer) are connected to GND layers only with 3 vias - its not enough for 30A current! I would not use thermal reliefs for those vias.

3. HS Mosfet gate traces change layers twice (increased inductance and reflections), they go far away from SHX traces what make inductance in HS Mosfet gate circuits very high (it works as a single turn coil), especially trace for phase B (Inner layer 1) goes left and takes very long way to DRV.

4. Decoupling caps 2,2uF between HS Mosfet drains and GND, there should be 3-4 (or more) GND vias around GND terminals of those caps (or even inside capacitor pads, then its called via in pad). Parasitic inductance between those caps and ground plane (or LS Mosfet sources) should me as low as possible.

5. DRV8353 - ground pad should be connected with thermal vias to ground plane/planes without thermal reliefs, DRV8353 can give quite a lot of heat and should be well cooled. 

Taking into account only 4 above mentioned flaws I can say it is quite a bad design. At first look the older working version is not much different. Making a board without those flaws doesn't guarantee proper working of circuit but it would be a good start to solve any potential problems. I think you have a few options:

- try to make work new boards with some simple fixes, it might be possible

- try to use older design, possibly make some improvements, an easier option

- design a new board preferably based on some working example like evm board. I think it would be the best option. With currents of 30A I think you would be able to use non-splitted ground design like two designs above (for higher currents splitted ground design may be needed, but I don't have experience in that area).

There are a few rules that I try to stick to:

- keep traces between H-bridge and DRV short and quite wide, do as little layer changes as possible

- keep SHX and GHX traces close together to minimize inductance (they form something like differential pair)

- keep as many traces as possible on the same layer as DRV, DRV ICs usually are designed to facilitate such routing

- keep low signal traces for DRV far away from its high power traces, DRV ICs usually have separate sides for low power and high power terminals

- use inner layer closer to Mosfets as ground plane with as little holes as possible, it helps to spread heat from traces on Mosfet layers

- use another inner layer as ground plane as much as it is possible, again better heat transfer and it provides reference ground for traces that changes layers

Once you decide how to proceed with your design I can try to help you as much as I can.

Michael may give you some answers as well.

Regards,

Grzegorz 

Grzegorz,

There indeed is one overly long VG trace on the B phase which we have bypassed with a wire, however, without any positive results. The rest of them are in the same order of magnitude as the ones on the working PCB. Do you recommend doing this on all GHx and SHx traces? 

Kr,

Alexander

Hi Alexander & Grzegorz,

After reviewing with the team, we believe the root cause of this is a layout issue. See Grzegorz post above as he made several good points (specifically points 2,3,4).

Have you tried swapping the FETS from the old board to the new board & to see if the fault condition goes away?

There are not many vias under the sense resistor, we suggest that you add more of them if your next design if possible.

However since the old board worked fine, and had the same amount of vias, this has us confused.

Let us know how you plan to proceed after trying to implement some of our suggestions. 

Thanks,

Michael

Hi Michael & Grzegorz,

Thank you for your input!

We have also looked at the possibility of a layout issue and have tried bypassing the Rsense GND vias with thick wires to a ground plane and bypassing the VGx with short wire jumpers as well. This had no effect. 

Moreover, is it possible that we won't be able to see the impact of these layout flaws in our design while measuring? 

Something that keeps bugging me is the little spike on the falling edge of the lo-side MOSFETs which we keep seeing even if we disconnect the FET and measure on the wire soldered directly to the pad/trace of the DRV. 

Michael, 

I would love to try swapping the FETs, however it's just not possible with the type of foil I am using. Do you thinkg bypassing all these poor connections with wire jumpers might solve the issue?

Also, we have tweaked our RCD-clamp and now our most powerful motor seems to work fine, it's the smaller ones that are acting up now. 

Kr,

Alexander

Alexander,

Rsense GND vias - their number is to small for 30A current, they may work for a while with no problems and then burn or give a lot of heat in the best case scenario, 10-15 of 0.3-0.4mm vias should do the work.

"bypassing the VGx with short wire jumpers as well" - I guess you meant GHx and SHx traces, it may help or not (if that is a source of problems). Short trace is not everything, it should be well coupled with its return path, so I think GHx and SHx traces/wires should be close to each other. After adding wires it may be necessary to cut old traces at both ends.

 "is it possible that we won't be able to see the impact of these layout flaws in our design while measuring?" - I think it is quite possible; bad H-bridge power supply decoupling, influence of parasitic inductances in H-bridge itself can be easily observed while motor current is high. In case of Mosfet gate circuits I guess it will be more difficult (I have not experienced problems with layout of gate circuits myself yet). Long gate traces mean high inductances what mean high Q of possible resonant LC circuits (mosfets add C to those circuits), resonance may appear under some conditions and then disappear, probably you would need to monitor both ends of trace to be sure that everything is OK. Long badly routed (far away from return paths or/and changing layers with return path badly designed) traces act as antennas receiving noise as well propagating noise. I think the best way is to route gate traces the best as it is possible and hope that it is good enough.

"Something that keeps bugging me is the little spike on the falling edge of the lo-side MOSFETs which we keep seeing even if we disconnect the FET and measure on the wire soldered directly to the pad/trace of the DRV." - if you could provide some more details, what pin of DRV, what exactly is disconnected, is driver working, is there any current flowing to motor, waveform would be helpful as well.

Regards,

Grzegorz

Grzegorz,

Great suggestions on all of the posts above. Thanks for your input, I appreciate the time you're putting into this.

Alexander,

If it's not a big deal to implement this I say go for it (bypassing all the bad connections with wire jumpers). Give it a shot, and let us know your results.

Did I understand this right? After you adjusted the RCD clamp the fault stops occurring on phase B with the more powerful motor? Is this starting the motor at 56V and not ramping it from lower voltage?

I look forward to hearing back, this is an interesting problem you're having. 

Regards,

Michael

Hi Michael,

So, we have tried connecting the shunts lo-sides to GND and bypassing the gate-traces separately, never all of them together. Neither measure yielded any positive results. Do you recon it's worth trying all of them together?

And yes, after adjusting the RCD-clamp to 330r and 22nF we could complete a first 10hr long current control durability test with a loaded traction motor with very clean current and voltage measurements (which I can provide iuw). However the steering motor (the lighter one) still instantly faults.

Kr,

Alexander

Hi Alexander,

Could you try to bypass all 100R resistors in current sense filters?

Regards,

Grzegorz

Hi Alexander,

I would suggest trying all of them together. Good I'm glad at least one of the motors is working with no issues.

Given that you are running two different motors you may have to tune the RCD clamp for each motor. Potentially try playing around with the IDRIVE settings for the new motor?

I'm interested to see the good waveforms from the traction motor.

Thanks,

Michael

Hi Alexander,

Have you had any luck getting getting your steering motor to spin?

Is there any other assistance I can provide? 

Thanks,

Michael

Hi Alexander,

I am going to close this thread, but feel free to ask a new question!

Thanks,

Matt